|
1 | 1 | from typing import Any, Dict, List |
2 | 2 | from datasets import load_dataset |
3 | | -from ...smp import * |
4 | 3 | from ..text_base import TextBaseDataset |
5 | 4 | from ..utils.judge_util import * |
6 | 5 | import concurrent.futures |
7 | 6 | import requests |
8 | 7 | import shutil |
9 | 8 | import ast |
| 9 | +from ...smp.file import dump, load ,get_intermediate_file_path |
10 | 10 | from openai import OpenAI |
11 | 11 | from json_repair import repair_json |
| 12 | +import pandas as pd |
| 13 | +import time |
| 14 | +import subprocess |
12 | 15 |
|
13 | 16 | save_dir = "./outputs/sgi_code_logs" |
14 | 17 | tmp_data_dir = "./outputs/sgi_tmp_data" |
@@ -384,10 +387,10 @@ def evaluate(self, eval_file, **judge_kwargs): |
384 | 387 | os.makedirs(os.path.join(tmp_data_dir, "0200"), exist_ok=True) |
385 | 388 | os.makedirs(os.path.join(tmp_data_dir, "0236"), exist_ok=True) |
386 | 389 |
|
387 | | - # download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/t10k-images-idx3-ubyte.gz", tmp_data_dir+"/0206") |
388 | | - # download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/t10k-labels-idx1-ubyte.gz", tmp_data_dir+"/0206") |
389 | | - # download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/train-images-idx3-ubyte.gz", tmp_data_dir+"/0206") |
390 | | - # download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/train-labels-idx1-ubyte.gz", tmp_data_dir+"/0206") |
| 390 | + download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/t10k-images-idx3-ubyte.gz", tmp_data_dir+"/0206") |
| 391 | + download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/t10k-labels-idx1-ubyte.gz", tmp_data_dir+"/0206") |
| 392 | + download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/train-images-idx3-ubyte.gz", tmp_data_dir+"/0206") |
| 393 | + download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0206/train-labels-idx1-ubyte.gz", tmp_data_dir+"/0206") |
391 | 394 |
|
392 | 395 | download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0200/adult.data", tmp_data_dir+"/0200") |
393 | 396 | download_file("https://raw.githubusercontent.com/InternScience/SGI-Bench/main/evaluation/task_3_dry_experiment/data/SGI_DryExperiment_0200/adult.test", tmp_data_dir+"/0200") |
|
0 commit comments